The DEAP Collaboration, a group of over 65 researchers from 18 institutions in Canada, the UK, Mexico, Germany, Spain, and Russia has developed the DEAP-3600 detector – one of the most sensitive experiments for the direct detection of dark matter.

DEAP has a competitive sensitivity to WIMP dark matter due to the design of the detector, where all components have been selected for extremely low levels of radioactivity, the very large target mass possible with liquid argon, and the very low background level achievable at the unique SNOLAB facility – a deep underground laboratory which uses the 2 km of rock overburden to filter out cosmic-ray muons and associated backgrounds that would otherwise interfere with WIMP detection.

At CIEMAT we are mainly involved in the efforts to mitigate the effects of background in dark matter searches using deep learning techniques, such as multilayer perceptrons and convolutional neural networks.

 

Currently, we are integrating into the main analysis pipeline an algorithm we have developed, a Multilayer Perceptron (MLP) that can improve the background rejection of alpha backgrounds coming from the neck of the detector. The incorporation of this algorithm results in an improvement in WIMP-like signal acceptance by a factor larger than two. The MLP used the proportion of light seen by each of the 255 photomultiplier tubes in each event to classify between nuclear recoils and this source of backgrounds. The events used for training and testing were simulated using Geant4, with a detailed geometry and optical model of the detector. The applicability of the classifier to real data was validated comparing the performance for the abundant 39Ar events both in data and simulation.